Surrogate Model for Feasible Process Control Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods of process optimization often generate recommendations for set points that are infeasible to implement due to process limitations, and fail to accommodate dynamic business requirements.
Innovation Solution
A computer-implemented system that utilizes action trajectories to train a local action surrogate model with an infeasible action penalty, enabling the system to provide tailored recommendations for process control that are feasible and optimal.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional process optimization methods are used, then mathematical optimization can be achieved, but the recommendations become infeasible due to process limitations
Solution Approach 1:
The patent changes the parameters of the optimization problem by incorporating real-time operation constraints and dynamic business requirements into the optimization formulation. The surrogate model learns from historical data with embedded constraints and generates recommendations that satisfy both mathematical optimality and practical feasibility, resolving the contradiction between optimization precision and implementation feasibility.
Solution Approach 2:
The system implements feedback by continuously monitoring real-time operation constraints and dynamic business requirements, then feeding this information back into the optimization formulation. The surrogate model uses historical data with embedded constraints to generate recommendations, creating a closed-loop system that ensures feasibility while maintaining optimization precision.
2Manufacturing precision
If static optimization models are used, then mathematical solutions can be obtained, but the system fails to adapt to dynamic business requirements
Solution Approach 1:
The patent applies dynamics by transitioning from static optimization models to a dynamic system that continuously adapts to changing business requirements and operation constraints. The surrogate model is trained on historical data that includes temporal patterns, and the optimization formulation incorporates real-time constraints, enabling the system to maintain optimization accuracy while adapting to dynamic requirements.
Solution Approach 2:
The system performs preliminary action by pre-training the surrogate model on historical data that captures temporal patterns and operational constraints before new optimization problems arise. This preliminary training enables the model to quickly adapt to dynamic business requirements without requiring complete retraining, maintaining both accuracy and adaptability.
3Manufacturing precision
If complex optimization formulations are used, then optimal solutions can be found, but the system becomes computationally intensive and slow
Solution Approach 1:
The patent uses copying by creating a surrogate model that replicates the complex optimization formulation's behavior but with reduced computational complexity. The surrogate model is trained on historical data and captures the essential relationships and constraints, allowing it to generate optimal recommendations much faster than the original complex formulation without sacrificing optimality.
Solution Approach 2:
The system employs a computationally inexpensive surrogate model that can be quickly trained on historical data and then used repeatedly for optimization. This disposable-like approach replaces the need for repeated expensive complex optimization calculations, significantly reducing computational time while maintaining optimization quality through the surrogate model's learned patterns.
Data Source
AI summary
Systems/techniques that facilitate tailored recommendation for process control by capturing real-time operation practices are provided. In various embodiments, a system can comprise a learning component that can employ a VQ-VAE based generative model to learn correlated patterns of state and control variables. In various embodiments, the system can further comprise a training component that can generate, based on the learned correlated patterns, feasible and infeasible action profiles to produce an infeasible and feasible system response. Furthermore, the feasible action profiles and system responses can be used with an infeasible action penalty to train a surrogate model, from which an analysis component can use to provide a recommendation of feasible and optimal set points of control variables.